Re: [PATCH v4] clocksource: arch_timer: Allow the device tree to specify uninitialized timer registers

[Yingjoe Chen <yingjoe.chen@mediatek.com>]

Hi,

On Wed, 2014-11-26 at 08:14 -0800, Doug Anderson wrote:
> Yingjoe,
> 
> On Wed, Nov 26, 2014 at 6:41 AM, Yingjoe Chen <yingjoe.chen@mediatek.com> wrote:
> > Sorry for the (very) late reply.
> > I just realize today MT8135 need this and the other patch [1] to boot
> > SMP correctly. I've applied both patches and they works fine. Thanks :)
> 
> Excellent.  It's helpful to include a Tested-by: tag in your email.
> You'd have a line with just "Tested-by: Yingjoe Chen
> <yingjoe.chen@mediatek.com>"

sure, here's my tested-by for the 2 patches

Tested-by: Yingjoe Chen <yingjoe.chen@mediatek.com>

I'll remember to add it next time :)


> > However, I'm not sure if we really need to add new property.
> > arm_arch_timer driver will only use virtual timer when virtual PPI
> > interrupt is provided, so the following patch to timer dtsi will also
> > works. I think if the firmware doesn't support virtual timer, it make
> > sense to not supply virtual interrupt.
> >
> >         timer {
> >                 compatible = "arm,armv7-timer";
> >                 interrupts = <GIC_PPI 13 IRQ_TYPE_LEVEL_LOW>,
> > -                            <GIC_PPI 14 IRQ_TYPE_LEVEL_LOW>,
> > -                            <GIC_PPI 11 IRQ_TYPE_LEVEL_LOW>,
> > -                            <GIC_PPI 10 IRQ_TYPE_LEVEL_LOW>;
> > +                            <GIC_PPI 14 IRQ_TYPE_LEVEL_LOW>;
> >                 clock-frequency = <13000000>;
> >         };
> 
> Once you have Sonny's patch then I believe that the above would work.
> However we rejected something like this because device tree is
> supposed to describe the hardware.  The hardware really does provide
> the virtual timer interrupts and they really are at PPI 11 and PPI 10.
> It's just that firmware doesn't handle things properly so they can't
> be used.
> 
> NOTE: If we add the "arm,cpu-registers-not-fw-configured" to the
> device tree and firmware actually works out how to configure things
> (like if somehow has firmware that has a hypervisor) then it can
> easily remove this device tree property before calling through to the
> kernel.  It would be much harder for the firmware to add back in the
> "PPI 11" and "PPI 10" entries to the timer.
> 
> -Doug

I see your point, that's good for me then.
Thanks.

Joe.C